基于深度学习的手部穴位定位研究  

Research on Hand Acupoint Location Based on Deep Learning

作　　者：常志强 张杰[1] 彭科 李宏[1] CHANGE Zhiqiang;ZHANG Jie;PENG Ke;LI Hong(College of Automation(Artificial Intelligence),Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)

机构地区：[1]杭州电子科技大学自动化(人工智能)学院,浙江杭州310018

出　　处：《传感技术学报》2024年第11期1893-1902,共10页Chinese Journal of Sensors and Actuators

摘　　要：提出一种基于深度学习的人体手部穴位定位方法,实现完整手部经脉寻找、穴位定位的解决方案。选择以ResNet152作为骨干网络,采用热图回归法进行穴位的定位,并引入经脉进行穴位的寻找,从而提高穴位检测模型的精度;将U-Net网络加入模型排除不同背景对实验的影响,增强了模型的定位精度和适用范围;引入注意力机制来使得模型更加关注相关特征,增强网络对有效特征的学习能力;通过实际手部照片为输入,进行模型准确度验证。模型能够识别手部的6条经脉和11个常用穴位,数据集中图像的平均准确率为93.6%;在自采集照片验证实验中,平均准确率为88.5%。该网络模型在手部穴位定位中有较好的准确率,为开发按摩机器人提供理论依据。Based on deep learning,a positioning method of human hand acupoint is proposed to realize a complete hand meridian search and acupoint location.ResNet152 is selected as the backbone network,heat map regression method is used to locate acupoint,and meridians are introduced to find acupoint,so as to improve the accuracy of acupoint detection model.To enhance the positioning accuracy and application scope of the model,the U-Net network is applied,which can eliminate the influence of different backgrounds on the experiment.The introduction of attention mechanism makes the model pay more attention to relevant features and enhances the learning ability of the network for effective features.Model accuracy is verified by inputting actual hand photos.The model could identify 6 meridians and 11 common hand acupoints,and the average accuracy of the images is 93.6%in the dataset and 88.5%in the verification experiment of self-acquisition photos.The network model has a better accuracy in hand acupoint location,which provides a theoretical basis for the development of massage robots.

关 键 词：经脉穴位 深度学习 关键点定位 图像处理 ResNet152 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]